Bridging the day and night domain gap for semantic segmentation
Authors
Romera Carmena, Eduardo; Bergasa Pascual, Luis Miguel; Yang, Kailun; Alvarez, Jose M.; Barea Navarro, RafaelIdentifiers
Permanent link (URI): http://hdl.handle.net/10017/45110DOI: 10.1109/IVS.2019.8813888
ISBN: 978-1-7281-0561-1
Publisher
IEEE
Date
2019-06Funders
Ministerio de Economía y competitividad
Comunidad de Madrid
Bibliographic citation
Romera, E., Bergasa, L. M., Yang, K., Álvarez, J. M. & Barea, R. 2019, "Bridging the day and night domain gap for semantic segmentation", en 2019 IEEE Intelligent Vehicles Symposium (IV), Paris, France, 2019, pp. 1312-1318
Description / Notes
2019 IEEE Intelligent Vehicles Symposium (IV), Paris, France, 9-12 Jun. 2019
Project
info:eu-repo/grantAgreement/MINECO//TRA2015-70501-C2-1-R/ES/VEHICULO INTELIGENTE PARA PERSONAS MAYORES/
info:eu-repo/grantAgreement/CAM//P2018%2FNMT-4331/ES/Madrid Robotics Digital Innovation Hub/RoboCity2030-DIH-CM
Document type
info:eu-repo/semantics/conferenceObject
Version
info:eu-repo/semantics/acceptedVersion
Publisher's version
https://doi.org/10.1109/IVS.2019.8813888Rights
Attribution-NonCommercial-NoDerivatives 4.0 Internacional
© 2019 IEEE
Access rights
info:eu-repo/semantics/openAccess
Abstract
Perception in autonomous vehicles has progressed
exponentially in the last years thanks to the advances of visionbased methods such as Convolutional Neural Networks (CNNs).
Current deep networks are both efficient and reliable, at least
in standard conditions, standing as a suitable solution for the
perception tasks of autonomous vehicles. However, there is
a large accuracy downgrade when these methods are taken
to adverse conditions such as nighttime. In this paper, we
study methods to alleviate this accuracy gap by using recent
techniques such as Generative Adversarial Networks (GANs).
We explore diverse options such as enlarging the dataset to
cover these domains in unsupervised training or adapting the
images on-the-fly during inference to a comfortable domain
such as sunny daylight in a pre-processing step. The results
show some interesting insights and demonstrate that both
proposed approaches considerably reduce the domain gap,
allowing IV perception systems to work reliably also at night.
Files in this item
Files | Size | Format |
|
---|---|---|---|
Bridging_Romera_IVS_2019.pdf | 3.898Mb |
|
Files | Size | Format |
|
---|---|---|---|
Bridging_Romera_IVS_2019.pdf | 3.898Mb |
|